Approximate Polygon Recovery from Rasterised Footprints

Geometrically regular building outlines are essential for urban analytics, geographic information systems, and spatial modelling. However, conventional regularisation techniques presuppose explicit vector geometry that rasterised products do not provide, rendering direct polygon-level correction ill-posed in raster space. To address this gap, this work presents a hybrid GeoAI framework that treats footprint regularisation from rasterised data as an approximate polygon recovery problem. Mask R-CNN first produces building masks, which are enhanced by three complementary operators: RANSAC thresholding for topological integrity, rectangular regularisation for orthogonal alignment, and fuzzy edge refinement for boundary de-noising. Analysing urban settlements, population density, updating the geographical records, and many various other factors does get affected by the spatial distributions of buildings. These footprints not only require semantic cor-rectness but also geometric regularity, including straight boundaries, orthogonal corners, and topological consistency. Developing precise and reliable building extraction techniques has become a significant and challenging research issue which is receiving more attention because of the diversity of buildings (e.g., in color, shape, size, materials, etc.) in multiple locations and the compa-rable nature of buildings to the background or additional objects.